How AI improves IT project management
IT project management has always stumbled over the same difficulties: optimistic estimates, risks detected too late, time-consuming reporting. Artificial intelligence is now built into project management tools and brings concrete answers to these points of friction, not by replacing the project manager, but by giving them a head start.
The findings behind this shift
Industry figures show that the problem AI sets out to address is far from marginal. Only 29% of companies complete their projects without any delay, and only 43% say they stay within the budget initially allocated. In other words, in most IT projects something goes off course along the way, often for lack of anticipation rather than lack of skill.
Projects completed without any delay: 29%
Projects completed within the initial budget: 43%
Distributed teams using at least one AI coordination tool: 74%
Time saved per week and per team thanks to AI automation: 8 to 12 hours
Gap between committed and booked budget on digital transformation projects: 15 to 20%
The idea is not to hand the project over to a machine, but to delegate what is repetitive or analytical in order to free up time for what calls for human judgement. Here is where AI brings real value.
What AI concretely brings
Predictive planning: by analysing the history of past projects, AI refines lead time estimates and flags unrealistic schedules before they slip, directly addressing the cause of the 71% of projects that run late.
Risk detection: it spots weak signals (accumulating delays, an overloaded team, critical dependencies) that are often invisible in the daily flow.
Automation: meeting minutes, task assignment, automatic reminders. Added together, these administrative tasks represent 8 to 12 hours per week and per team according to the most recent studies.
Decision support: dashboards enriched with contextual recommendations, to decide faster on a factual basis.
What separates a good prediction from a bad one
Not all predictive planning features are equal. A reliable prediction relies on a history of at least several comparable projects (same type of work, similar team, close complexity), and is re-assessed continuously as the project progresses rather than staying frozen on the initial estimate. A prediction based on a history that is too thin or too heterogeneous (a mix of unrelated projects) produces margins of error that can far exceed those of an experienced human estimate, which is why data quality remains the decisive factor, even before the choice of tool.
Tools to watch
GitHub Copilot and its assisted workspaces: planning and carrying out development tasks with AI support.
Notion AI: automatic drafting and structuring of project documentation, often neglected for lack of time.
Jira and Linear with AI features: backlog prioritisation and effort estimation.
The market moves fast and the offering is constantly renewed; what matters is less the tool than the method it fits into. A bad process equipped with AI is still a bad process.
The conditions for success
Project management AI only gives good results if it is fed reliable data. Poorly completed tickets, an incomplete history or fanciful estimates will produce equally fanciful predictions. The quality of project data is the real prerequisite, which partly explains why the gap between committed and booked budget can still reach 15 to 20% even on well-equipped projects: AI amplifies the quality of existing data, it does not make up for its absence. Teams must also be trained, and what the tool decides or does not decide must be made clear, to avoid both mistrust and blind delegation.
AI does not replace the project manager
AI is a co-pilot, not a pilot. Human added value remains intact on what lies at the heart of the job: communication with stakeholders, negotiating trade-offs, managing tensions and making strategic decisions in an uncertain context. AI lights the way; it is always a human who chooses the direction. This complementarity echoes what we observe in auditing and optimising the information system more generally.
An example of gradual adoption
On a typical IT project, a reasonable introduction of project management AI starts with a single measurable use case, for example automating meeting minutes and reminders for overdue tasks. Once this gain has been validated and accepted by the team, the next step is to connect the history of past projects to refine lead time estimates for the next sprints. Extending to risk detection last, once the baseline data is reliable, avoids building predictions on a history that is still too thin.
The more projects, the more useful AI becomes
Project management AI is not equally effective for everyone, for a simple reason: it learns from the past. Its predictions of lead times and risks feed on the history of previous projects, so the more comparable projects an organisation runs, the more relevant the recommendations become. A company that carries out one assignment of the same type after another accumulates valuable raw material; AI finds recurring patterns in it and refines its estimates with each new project.
On a one-off project, with nothing comparable in the history, the benefit shrinks accordingly: the predictive part lacks references to be reliable. What remains is the automation of administrative tasks (minutes, reminders, follow-up), which is useful whatever the volume. In other words, project management AI pays off mainly through duration and repetition, not on a single shot.
Start with a single use case
Faced with these promises, the temptation is to switch everything on at once: prediction, risk detection, smart dashboards. That is often the best way to lose the team along the way and discredit the tool. A successful introduction, on the contrary, starts with a single measurable, low-risk use case, typically automating meeting minutes and reminders for overdue tasks. The gain is immediate and visible, and it does not depend on the quality of the history.
Once this first benefit has been validated and accepted, the history of past projects is connected to refine estimates, and only then, last of all, is risk detection switched on, when the baseline data is finally reliable. This gradual ramp-up avoids building predictions on a history that is still too thin, and gives the team time to get to grips with the tool rather than having it forced on them.
A tool is only as good as the method
In the end, project management AI works no miracles: it amplifies what already exists. A team that records its projects poorly will get fanciful predictions; a rigorous team will see its estimates become more reliable and its time freed up. The real success factor is therefore not the choice of tool but the quality of the data entrusted to it and the method it fits into. Start small, feed the tool clean data and keep humans in charge of decisions: that is what separates a real gain from an expensive gadget.
We use these tools in our IT project management assignments, in the service of the method rather than in its place. Contact us for more precise tracking and better anticipation of risks.
Sources: Les Makers, project management statistics 2026 (in French) · PMO at Work, project management and AI in 2026 (in French)